xref: /openbmc/linux/arch/ia64/kernel/setup.c (revision bc5aa3a0)
1 /*
2  * Architecture-specific setup.
3  *
4  * Copyright (C) 1998-2001, 2003-2004 Hewlett-Packard Co
5  *	David Mosberger-Tang <davidm@hpl.hp.com>
6  *	Stephane Eranian <eranian@hpl.hp.com>
7  * Copyright (C) 2000, 2004 Intel Corp
8  * 	Rohit Seth <rohit.seth@intel.com>
9  * 	Suresh Siddha <suresh.b.siddha@intel.com>
10  * 	Gordon Jin <gordon.jin@intel.com>
11  * Copyright (C) 1999 VA Linux Systems
12  * Copyright (C) 1999 Walt Drummond <drummond@valinux.com>
13  *
14  * 12/26/04 S.Siddha, G.Jin, R.Seth
15  *			Add multi-threading and multi-core detection
16  * 11/12/01 D.Mosberger Convert get_cpuinfo() to seq_file based show_cpuinfo().
17  * 04/04/00 D.Mosberger renamed cpu_initialized to cpu_online_map
18  * 03/31/00 R.Seth	cpu_initialized and current->processor fixes
19  * 02/04/00 D.Mosberger	some more get_cpuinfo fixes...
20  * 02/01/00 R.Seth	fixed get_cpuinfo for SMP
21  * 01/07/99 S.Eranian	added the support for command line argument
22  * 06/24/99 W.Drummond	added boot_cpu_data.
23  * 05/28/05 Z. Menyhart	Dynamic stride size for "flush_icache_range()"
24  */
25 #include <linux/module.h>
26 #include <linux/init.h>
27 
28 #include <linux/acpi.h>
29 #include <linux/bootmem.h>
30 #include <linux/console.h>
31 #include <linux/delay.h>
32 #include <linux/kernel.h>
33 #include <linux/reboot.h>
34 #include <linux/sched.h>
35 #include <linux/seq_file.h>
36 #include <linux/string.h>
37 #include <linux/threads.h>
38 #include <linux/screen_info.h>
39 #include <linux/dmi.h>
40 #include <linux/serial.h>
41 #include <linux/serial_core.h>
42 #include <linux/efi.h>
43 #include <linux/initrd.h>
44 #include <linux/pm.h>
45 #include <linux/cpufreq.h>
46 #include <linux/kexec.h>
47 #include <linux/crash_dump.h>
48 
49 #include <asm/machvec.h>
50 #include <asm/mca.h>
51 #include <asm/meminit.h>
52 #include <asm/page.h>
53 #include <asm/patch.h>
54 #include <asm/pgtable.h>
55 #include <asm/processor.h>
56 #include <asm/sal.h>
57 #include <asm/sections.h>
58 #include <asm/setup.h>
59 #include <asm/smp.h>
60 #include <asm/tlbflush.h>
61 #include <asm/unistd.h>
62 #include <asm/hpsim.h>
63 
64 #if defined(CONFIG_SMP) && (IA64_CPU_SIZE > PAGE_SIZE)
65 # error "struct cpuinfo_ia64 too big!"
66 #endif
67 
68 #ifdef CONFIG_SMP
69 unsigned long __per_cpu_offset[NR_CPUS];
70 EXPORT_SYMBOL(__per_cpu_offset);
71 #endif
72 
73 DEFINE_PER_CPU(struct cpuinfo_ia64, ia64_cpu_info);
74 DEFINE_PER_CPU(unsigned long, local_per_cpu_offset);
75 unsigned long ia64_cycles_per_usec;
76 struct ia64_boot_param *ia64_boot_param;
77 struct screen_info screen_info;
78 unsigned long vga_console_iobase;
79 unsigned long vga_console_membase;
80 
81 static struct resource data_resource = {
82 	.name	= "Kernel data",
83 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
84 };
85 
86 static struct resource code_resource = {
87 	.name	= "Kernel code",
88 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
89 };
90 
91 static struct resource bss_resource = {
92 	.name	= "Kernel bss",
93 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
94 };
95 
96 unsigned long ia64_max_cacheline_size;
97 
98 unsigned long ia64_iobase;	/* virtual address for I/O accesses */
99 EXPORT_SYMBOL(ia64_iobase);
100 struct io_space io_space[MAX_IO_SPACES];
101 EXPORT_SYMBOL(io_space);
102 unsigned int num_io_spaces;
103 
104 /*
105  * "flush_icache_range()" needs to know what processor dependent stride size to use
106  * when it makes i-cache(s) coherent with d-caches.
107  */
108 #define	I_CACHE_STRIDE_SHIFT	5	/* Safest way to go: 32 bytes by 32 bytes */
109 unsigned long ia64_i_cache_stride_shift = ~0;
110 /*
111  * "clflush_cache_range()" needs to know what processor dependent stride size to
112  * use when it flushes cache lines including both d-cache and i-cache.
113  */
114 /* Safest way to go: 32 bytes by 32 bytes */
115 #define	CACHE_STRIDE_SHIFT	5
116 unsigned long ia64_cache_stride_shift = ~0;
117 
118 /*
119  * The merge_mask variable needs to be set to (max(iommu_page_size(iommu)) - 1).  This
120  * mask specifies a mask of address bits that must be 0 in order for two buffers to be
121  * mergeable by the I/O MMU (i.e., the end address of the first buffer and the start
122  * address of the second buffer must be aligned to (merge_mask+1) in order to be
123  * mergeable).  By default, we assume there is no I/O MMU which can merge physically
124  * discontiguous buffers, so we set the merge_mask to ~0UL, which corresponds to a iommu
125  * page-size of 2^64.
126  */
127 unsigned long ia64_max_iommu_merge_mask = ~0UL;
128 EXPORT_SYMBOL(ia64_max_iommu_merge_mask);
129 
130 /*
131  * We use a special marker for the end of memory and it uses the extra (+1) slot
132  */
133 struct rsvd_region rsvd_region[IA64_MAX_RSVD_REGIONS + 1] __initdata;
134 int num_rsvd_regions __initdata;
135 
136 
137 /*
138  * Filter incoming memory segments based on the primitive map created from the boot
139  * parameters. Segments contained in the map are removed from the memory ranges. A
140  * caller-specified function is called with the memory ranges that remain after filtering.
141  * This routine does not assume the incoming segments are sorted.
142  */
143 int __init
144 filter_rsvd_memory (u64 start, u64 end, void *arg)
145 {
146 	u64 range_start, range_end, prev_start;
147 	void (*func)(unsigned long, unsigned long, int);
148 	int i;
149 
150 #if IGNORE_PFN0
151 	if (start == PAGE_OFFSET) {
152 		printk(KERN_WARNING "warning: skipping physical page 0\n");
153 		start += PAGE_SIZE;
154 		if (start >= end) return 0;
155 	}
156 #endif
157 	/*
158 	 * lowest possible address(walker uses virtual)
159 	 */
160 	prev_start = PAGE_OFFSET;
161 	func = arg;
162 
163 	for (i = 0; i < num_rsvd_regions; ++i) {
164 		range_start = max(start, prev_start);
165 		range_end   = min(end, rsvd_region[i].start);
166 
167 		if (range_start < range_end)
168 			call_pernode_memory(__pa(range_start), range_end - range_start, func);
169 
170 		/* nothing more available in this segment */
171 		if (range_end == end) return 0;
172 
173 		prev_start = rsvd_region[i].end;
174 	}
175 	/* end of memory marker allows full processing inside loop body */
176 	return 0;
177 }
178 
179 /*
180  * Similar to "filter_rsvd_memory()", but the reserved memory ranges
181  * are not filtered out.
182  */
183 int __init
184 filter_memory(u64 start, u64 end, void *arg)
185 {
186 	void (*func)(unsigned long, unsigned long, int);
187 
188 #if IGNORE_PFN0
189 	if (start == PAGE_OFFSET) {
190 		printk(KERN_WARNING "warning: skipping physical page 0\n");
191 		start += PAGE_SIZE;
192 		if (start >= end)
193 			return 0;
194 	}
195 #endif
196 	func = arg;
197 	if (start < end)
198 		call_pernode_memory(__pa(start), end - start, func);
199 	return 0;
200 }
201 
202 static void __init
203 sort_regions (struct rsvd_region *rsvd_region, int max)
204 {
205 	int j;
206 
207 	/* simple bubble sorting */
208 	while (max--) {
209 		for (j = 0; j < max; ++j) {
210 			if (rsvd_region[j].start > rsvd_region[j+1].start) {
211 				struct rsvd_region tmp;
212 				tmp = rsvd_region[j];
213 				rsvd_region[j] = rsvd_region[j + 1];
214 				rsvd_region[j + 1] = tmp;
215 			}
216 		}
217 	}
218 }
219 
220 /* merge overlaps */
221 static int __init
222 merge_regions (struct rsvd_region *rsvd_region, int max)
223 {
224 	int i;
225 	for (i = 1; i < max; ++i) {
226 		if (rsvd_region[i].start >= rsvd_region[i-1].end)
227 			continue;
228 		if (rsvd_region[i].end > rsvd_region[i-1].end)
229 			rsvd_region[i-1].end = rsvd_region[i].end;
230 		--max;
231 		memmove(&rsvd_region[i], &rsvd_region[i+1],
232 			(max - i) * sizeof(struct rsvd_region));
233 	}
234 	return max;
235 }
236 
237 /*
238  * Request address space for all standard resources
239  */
240 static int __init register_memory(void)
241 {
242 	code_resource.start = ia64_tpa(_text);
243 	code_resource.end   = ia64_tpa(_etext) - 1;
244 	data_resource.start = ia64_tpa(_etext);
245 	data_resource.end   = ia64_tpa(_edata) - 1;
246 	bss_resource.start  = ia64_tpa(__bss_start);
247 	bss_resource.end    = ia64_tpa(_end) - 1;
248 	efi_initialize_iomem_resources(&code_resource, &data_resource,
249 			&bss_resource);
250 
251 	return 0;
252 }
253 
254 __initcall(register_memory);
255 
256 
257 #ifdef CONFIG_KEXEC
258 
259 /*
260  * This function checks if the reserved crashkernel is allowed on the specific
261  * IA64 machine flavour. Machines without an IO TLB use swiotlb and require
262  * some memory below 4 GB (i.e. in 32 bit area), see the implementation of
263  * lib/swiotlb.c. The hpzx1 architecture has an IO TLB but cannot use that
264  * in kdump case. See the comment in sba_init() in sba_iommu.c.
265  *
266  * So, the only machvec that really supports loading the kdump kernel
267  * over 4 GB is "sn2".
268  */
269 static int __init check_crashkernel_memory(unsigned long pbase, size_t size)
270 {
271 	if (ia64_platform_is("sn2") || ia64_platform_is("uv"))
272 		return 1;
273 	else
274 		return pbase < (1UL << 32);
275 }
276 
277 static void __init setup_crashkernel(unsigned long total, int *n)
278 {
279 	unsigned long long base = 0, size = 0;
280 	int ret;
281 
282 	ret = parse_crashkernel(boot_command_line, total,
283 			&size, &base);
284 	if (ret == 0 && size > 0) {
285 		if (!base) {
286 			sort_regions(rsvd_region, *n);
287 			*n = merge_regions(rsvd_region, *n);
288 			base = kdump_find_rsvd_region(size,
289 					rsvd_region, *n);
290 		}
291 
292 		if (!check_crashkernel_memory(base, size)) {
293 			pr_warning("crashkernel: There would be kdump memory "
294 				"at %ld GB but this is unusable because it "
295 				"must\nbe below 4 GB. Change the memory "
296 				"configuration of the machine.\n",
297 				(unsigned long)(base >> 30));
298 			return;
299 		}
300 
301 		if (base != ~0UL) {
302 			printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
303 					"for crashkernel (System RAM: %ldMB)\n",
304 					(unsigned long)(size >> 20),
305 					(unsigned long)(base >> 20),
306 					(unsigned long)(total >> 20));
307 			rsvd_region[*n].start =
308 				(unsigned long)__va(base);
309 			rsvd_region[*n].end =
310 				(unsigned long)__va(base + size);
311 			(*n)++;
312 			crashk_res.start = base;
313 			crashk_res.end = base + size - 1;
314 		}
315 	}
316 	efi_memmap_res.start = ia64_boot_param->efi_memmap;
317 	efi_memmap_res.end = efi_memmap_res.start +
318 		ia64_boot_param->efi_memmap_size;
319 	boot_param_res.start = __pa(ia64_boot_param);
320 	boot_param_res.end = boot_param_res.start +
321 		sizeof(*ia64_boot_param);
322 }
323 #else
324 static inline void __init setup_crashkernel(unsigned long total, int *n)
325 {}
326 #endif
327 
328 /**
329  * reserve_memory - setup reserved memory areas
330  *
331  * Setup the reserved memory areas set aside for the boot parameters,
332  * initrd, etc.  There are currently %IA64_MAX_RSVD_REGIONS defined,
333  * see arch/ia64/include/asm/meminit.h if you need to define more.
334  */
335 void __init
336 reserve_memory (void)
337 {
338 	int n = 0;
339 	unsigned long total_memory;
340 
341 	/*
342 	 * none of the entries in this table overlap
343 	 */
344 	rsvd_region[n].start = (unsigned long) ia64_boot_param;
345 	rsvd_region[n].end   = rsvd_region[n].start + sizeof(*ia64_boot_param);
346 	n++;
347 
348 	rsvd_region[n].start = (unsigned long) __va(ia64_boot_param->efi_memmap);
349 	rsvd_region[n].end   = rsvd_region[n].start + ia64_boot_param->efi_memmap_size;
350 	n++;
351 
352 	rsvd_region[n].start = (unsigned long) __va(ia64_boot_param->command_line);
353 	rsvd_region[n].end   = (rsvd_region[n].start
354 				+ strlen(__va(ia64_boot_param->command_line)) + 1);
355 	n++;
356 
357 	rsvd_region[n].start = (unsigned long) ia64_imva((void *)KERNEL_START);
358 	rsvd_region[n].end   = (unsigned long) ia64_imva(_end);
359 	n++;
360 
361 #ifdef CONFIG_BLK_DEV_INITRD
362 	if (ia64_boot_param->initrd_start) {
363 		rsvd_region[n].start = (unsigned long)__va(ia64_boot_param->initrd_start);
364 		rsvd_region[n].end   = rsvd_region[n].start + ia64_boot_param->initrd_size;
365 		n++;
366 	}
367 #endif
368 
369 #ifdef CONFIG_CRASH_DUMP
370 	if (reserve_elfcorehdr(&rsvd_region[n].start,
371 			       &rsvd_region[n].end) == 0)
372 		n++;
373 #endif
374 
375 	total_memory = efi_memmap_init(&rsvd_region[n].start, &rsvd_region[n].end);
376 	n++;
377 
378 	setup_crashkernel(total_memory, &n);
379 
380 	/* end of memory marker */
381 	rsvd_region[n].start = ~0UL;
382 	rsvd_region[n].end   = ~0UL;
383 	n++;
384 
385 	num_rsvd_regions = n;
386 	BUG_ON(IA64_MAX_RSVD_REGIONS + 1 < n);
387 
388 	sort_regions(rsvd_region, num_rsvd_regions);
389 	num_rsvd_regions = merge_regions(rsvd_region, num_rsvd_regions);
390 }
391 
392 
393 /**
394  * find_initrd - get initrd parameters from the boot parameter structure
395  *
396  * Grab the initrd start and end from the boot parameter struct given us by
397  * the boot loader.
398  */
399 void __init
400 find_initrd (void)
401 {
402 #ifdef CONFIG_BLK_DEV_INITRD
403 	if (ia64_boot_param->initrd_start) {
404 		initrd_start = (unsigned long)__va(ia64_boot_param->initrd_start);
405 		initrd_end   = initrd_start+ia64_boot_param->initrd_size;
406 
407 		printk(KERN_INFO "Initial ramdisk at: 0x%lx (%llu bytes)\n",
408 		       initrd_start, ia64_boot_param->initrd_size);
409 	}
410 #endif
411 }
412 
413 static void __init
414 io_port_init (void)
415 {
416 	unsigned long phys_iobase;
417 
418 	/*
419 	 * Set `iobase' based on the EFI memory map or, failing that, the
420 	 * value firmware left in ar.k0.
421 	 *
422 	 * Note that in ia32 mode, IN/OUT instructions use ar.k0 to compute
423 	 * the port's virtual address, so ia32_load_state() loads it with a
424 	 * user virtual address.  But in ia64 mode, glibc uses the
425 	 * *physical* address in ar.k0 to mmap the appropriate area from
426 	 * /dev/mem, and the inX()/outX() interfaces use MMIO.  In both
427 	 * cases, user-mode can only use the legacy 0-64K I/O port space.
428 	 *
429 	 * ar.k0 is not involved in kernel I/O port accesses, which can use
430 	 * any of the I/O port spaces and are done via MMIO using the
431 	 * virtual mmio_base from the appropriate io_space[].
432 	 */
433 	phys_iobase = efi_get_iobase();
434 	if (!phys_iobase) {
435 		phys_iobase = ia64_get_kr(IA64_KR_IO_BASE);
436 		printk(KERN_INFO "No I/O port range found in EFI memory map, "
437 			"falling back to AR.KR0 (0x%lx)\n", phys_iobase);
438 	}
439 	ia64_iobase = (unsigned long) ioremap(phys_iobase, 0);
440 	ia64_set_kr(IA64_KR_IO_BASE, __pa(ia64_iobase));
441 
442 	/* setup legacy IO port space */
443 	io_space[0].mmio_base = ia64_iobase;
444 	io_space[0].sparse = 1;
445 	num_io_spaces = 1;
446 }
447 
448 /**
449  * early_console_setup - setup debugging console
450  *
451  * Consoles started here require little enough setup that we can start using
452  * them very early in the boot process, either right after the machine
453  * vector initialization, or even before if the drivers can detect their hw.
454  *
455  * Returns non-zero if a console couldn't be setup.
456  */
457 static inline int __init
458 early_console_setup (char *cmdline)
459 {
460 	int earlycons = 0;
461 
462 #ifdef CONFIG_SERIAL_SGI_L1_CONSOLE
463 	{
464 		extern int sn_serial_console_early_setup(void);
465 		if (!sn_serial_console_early_setup())
466 			earlycons++;
467 	}
468 #endif
469 #ifdef CONFIG_EFI_PCDP
470 	if (!efi_setup_pcdp_console(cmdline))
471 		earlycons++;
472 #endif
473 	if (!simcons_register())
474 		earlycons++;
475 
476 	return (earlycons) ? 0 : -1;
477 }
478 
479 static inline void
480 mark_bsp_online (void)
481 {
482 #ifdef CONFIG_SMP
483 	/* If we register an early console, allow CPU 0 to printk */
484 	set_cpu_online(smp_processor_id(), true);
485 #endif
486 }
487 
488 static __initdata int nomca;
489 static __init int setup_nomca(char *s)
490 {
491 	nomca = 1;
492 	return 0;
493 }
494 early_param("nomca", setup_nomca);
495 
496 #ifdef CONFIG_CRASH_DUMP
497 int __init reserve_elfcorehdr(u64 *start, u64 *end)
498 {
499 	u64 length;
500 
501 	/* We get the address using the kernel command line,
502 	 * but the size is extracted from the EFI tables.
503 	 * Both address and size are required for reservation
504 	 * to work properly.
505 	 */
506 
507 	if (!is_vmcore_usable())
508 		return -EINVAL;
509 
510 	if ((length = vmcore_find_descriptor_size(elfcorehdr_addr)) == 0) {
511 		vmcore_unusable();
512 		return -EINVAL;
513 	}
514 
515 	*start = (unsigned long)__va(elfcorehdr_addr);
516 	*end = *start + length;
517 	return 0;
518 }
519 
520 #endif /* CONFIG_PROC_VMCORE */
521 
522 void __init
523 setup_arch (char **cmdline_p)
524 {
525 	unw_init();
526 
527 	ia64_patch_vtop((u64) __start___vtop_patchlist, (u64) __end___vtop_patchlist);
528 
529 	*cmdline_p = __va(ia64_boot_param->command_line);
530 	strlcpy(boot_command_line, *cmdline_p, COMMAND_LINE_SIZE);
531 
532 	efi_init();
533 	io_port_init();
534 
535 #ifdef CONFIG_IA64_GENERIC
536 	/* machvec needs to be parsed from the command line
537 	 * before parse_early_param() is called to ensure
538 	 * that ia64_mv is initialised before any command line
539 	 * settings may cause console setup to occur
540 	 */
541 	machvec_init_from_cmdline(*cmdline_p);
542 #endif
543 
544 	parse_early_param();
545 
546 	if (early_console_setup(*cmdline_p) == 0)
547 		mark_bsp_online();
548 
549 #ifdef CONFIG_ACPI
550 	/* Initialize the ACPI boot-time table parser */
551 	acpi_table_init();
552 	early_acpi_boot_init();
553 # ifdef CONFIG_ACPI_NUMA
554 	acpi_numa_init();
555 	acpi_numa_fixup();
556 #  ifdef CONFIG_ACPI_HOTPLUG_CPU
557 	prefill_possible_map();
558 #  endif
559 	per_cpu_scan_finalize((cpumask_weight(&early_cpu_possible_map) == 0 ?
560 		32 : cpumask_weight(&early_cpu_possible_map)),
561 		additional_cpus > 0 ? additional_cpus : 0);
562 # endif
563 #endif /* CONFIG_APCI_BOOT */
564 
565 #ifdef CONFIG_SMP
566 	smp_build_cpu_map();
567 #endif
568 	find_memory();
569 
570 	/* process SAL system table: */
571 	ia64_sal_init(__va(efi.sal_systab));
572 
573 #ifdef CONFIG_ITANIUM
574 	ia64_patch_rse((u64) __start___rse_patchlist, (u64) __end___rse_patchlist);
575 #else
576 	{
577 		unsigned long num_phys_stacked;
578 
579 		if (ia64_pal_rse_info(&num_phys_stacked, 0) == 0 && num_phys_stacked > 96)
580 			ia64_patch_rse((u64) __start___rse_patchlist, (u64) __end___rse_patchlist);
581 	}
582 #endif
583 
584 #ifdef CONFIG_SMP
585 	cpu_physical_id(0) = hard_smp_processor_id();
586 #endif
587 
588 	cpu_init();	/* initialize the bootstrap CPU */
589 	mmu_context_init();	/* initialize context_id bitmap */
590 
591 #ifdef CONFIG_VT
592 	if (!conswitchp) {
593 # if defined(CONFIG_DUMMY_CONSOLE)
594 		conswitchp = &dummy_con;
595 # endif
596 # if defined(CONFIG_VGA_CONSOLE)
597 		/*
598 		 * Non-legacy systems may route legacy VGA MMIO range to system
599 		 * memory.  vga_con probes the MMIO hole, so memory looks like
600 		 * a VGA device to it.  The EFI memory map can tell us if it's
601 		 * memory so we can avoid this problem.
602 		 */
603 		if (efi_mem_type(0xA0000) != EFI_CONVENTIONAL_MEMORY)
604 			conswitchp = &vga_con;
605 # endif
606 	}
607 #endif
608 
609 	/* enable IA-64 Machine Check Abort Handling unless disabled */
610 	if (!nomca)
611 		ia64_mca_init();
612 
613 	platform_setup(cmdline_p);
614 #ifndef CONFIG_IA64_HP_SIM
615 	check_sal_cache_flush();
616 #endif
617 	paging_init();
618 }
619 
620 /*
621  * Display cpu info for all CPUs.
622  */
623 static int
624 show_cpuinfo (struct seq_file *m, void *v)
625 {
626 #ifdef CONFIG_SMP
627 #	define lpj	c->loops_per_jiffy
628 #	define cpunum	c->cpu
629 #else
630 #	define lpj	loops_per_jiffy
631 #	define cpunum	0
632 #endif
633 	static struct {
634 		unsigned long mask;
635 		const char *feature_name;
636 	} feature_bits[] = {
637 		{ 1UL << 0, "branchlong" },
638 		{ 1UL << 1, "spontaneous deferral"},
639 		{ 1UL << 2, "16-byte atomic ops" }
640 	};
641 	char features[128], *cp, *sep;
642 	struct cpuinfo_ia64 *c = v;
643 	unsigned long mask;
644 	unsigned long proc_freq;
645 	int i, size;
646 
647 	mask = c->features;
648 
649 	/* build the feature string: */
650 	memcpy(features, "standard", 9);
651 	cp = features;
652 	size = sizeof(features);
653 	sep = "";
654 	for (i = 0; i < ARRAY_SIZE(feature_bits) && size > 1; ++i) {
655 		if (mask & feature_bits[i].mask) {
656 			cp += snprintf(cp, size, "%s%s", sep,
657 				       feature_bits[i].feature_name),
658 			sep = ", ";
659 			mask &= ~feature_bits[i].mask;
660 			size = sizeof(features) - (cp - features);
661 		}
662 	}
663 	if (mask && size > 1) {
664 		/* print unknown features as a hex value */
665 		snprintf(cp, size, "%s0x%lx", sep, mask);
666 	}
667 
668 	proc_freq = cpufreq_quick_get(cpunum);
669 	if (!proc_freq)
670 		proc_freq = c->proc_freq / 1000;
671 
672 	seq_printf(m,
673 		   "processor  : %d\n"
674 		   "vendor     : %s\n"
675 		   "arch       : IA-64\n"
676 		   "family     : %u\n"
677 		   "model      : %u\n"
678 		   "model name : %s\n"
679 		   "revision   : %u\n"
680 		   "archrev    : %u\n"
681 		   "features   : %s\n"
682 		   "cpu number : %lu\n"
683 		   "cpu regs   : %u\n"
684 		   "cpu MHz    : %lu.%03lu\n"
685 		   "itc MHz    : %lu.%06lu\n"
686 		   "BogoMIPS   : %lu.%02lu\n",
687 		   cpunum, c->vendor, c->family, c->model,
688 		   c->model_name, c->revision, c->archrev,
689 		   features, c->ppn, c->number,
690 		   proc_freq / 1000, proc_freq % 1000,
691 		   c->itc_freq / 1000000, c->itc_freq % 1000000,
692 		   lpj*HZ/500000, (lpj*HZ/5000) % 100);
693 #ifdef CONFIG_SMP
694 	seq_printf(m, "siblings   : %u\n",
695 		   cpumask_weight(&cpu_core_map[cpunum]));
696 	if (c->socket_id != -1)
697 		seq_printf(m, "physical id: %u\n", c->socket_id);
698 	if (c->threads_per_core > 1 || c->cores_per_socket > 1)
699 		seq_printf(m,
700 			   "core id    : %u\n"
701 			   "thread id  : %u\n",
702 			   c->core_id, c->thread_id);
703 #endif
704 	seq_printf(m,"\n");
705 
706 	return 0;
707 }
708 
709 static void *
710 c_start (struct seq_file *m, loff_t *pos)
711 {
712 #ifdef CONFIG_SMP
713 	while (*pos < nr_cpu_ids && !cpu_online(*pos))
714 		++*pos;
715 #endif
716 	return *pos < nr_cpu_ids ? cpu_data(*pos) : NULL;
717 }
718 
719 static void *
720 c_next (struct seq_file *m, void *v, loff_t *pos)
721 {
722 	++*pos;
723 	return c_start(m, pos);
724 }
725 
726 static void
727 c_stop (struct seq_file *m, void *v)
728 {
729 }
730 
731 const struct seq_operations cpuinfo_op = {
732 	.start =	c_start,
733 	.next =		c_next,
734 	.stop =		c_stop,
735 	.show =		show_cpuinfo
736 };
737 
738 #define MAX_BRANDS	8
739 static char brandname[MAX_BRANDS][128];
740 
741 static char *
742 get_model_name(__u8 family, __u8 model)
743 {
744 	static int overflow;
745 	char brand[128];
746 	int i;
747 
748 	memcpy(brand, "Unknown", 8);
749 	if (ia64_pal_get_brand_info(brand)) {
750 		if (family == 0x7)
751 			memcpy(brand, "Merced", 7);
752 		else if (family == 0x1f) switch (model) {
753 			case 0: memcpy(brand, "McKinley", 9); break;
754 			case 1: memcpy(brand, "Madison", 8); break;
755 			case 2: memcpy(brand, "Madison up to 9M cache", 23); break;
756 		}
757 	}
758 	for (i = 0; i < MAX_BRANDS; i++)
759 		if (strcmp(brandname[i], brand) == 0)
760 			return brandname[i];
761 	for (i = 0; i < MAX_BRANDS; i++)
762 		if (brandname[i][0] == '\0')
763 			return strcpy(brandname[i], brand);
764 	if (overflow++ == 0)
765 		printk(KERN_ERR
766 		       "%s: Table overflow. Some processor model information will be missing\n",
767 		       __func__);
768 	return "Unknown";
769 }
770 
771 static void
772 identify_cpu (struct cpuinfo_ia64 *c)
773 {
774 	union {
775 		unsigned long bits[5];
776 		struct {
777 			/* id 0 & 1: */
778 			char vendor[16];
779 
780 			/* id 2 */
781 			u64 ppn;		/* processor serial number */
782 
783 			/* id 3: */
784 			unsigned number		:  8;
785 			unsigned revision	:  8;
786 			unsigned model		:  8;
787 			unsigned family		:  8;
788 			unsigned archrev	:  8;
789 			unsigned reserved	: 24;
790 
791 			/* id 4: */
792 			u64 features;
793 		} field;
794 	} cpuid;
795 	pal_vm_info_1_u_t vm1;
796 	pal_vm_info_2_u_t vm2;
797 	pal_status_t status;
798 	unsigned long impl_va_msb = 50, phys_addr_size = 44;	/* Itanium defaults */
799 	int i;
800 	for (i = 0; i < 5; ++i)
801 		cpuid.bits[i] = ia64_get_cpuid(i);
802 
803 	memcpy(c->vendor, cpuid.field.vendor, 16);
804 #ifdef CONFIG_SMP
805 	c->cpu = smp_processor_id();
806 
807 	/* below default values will be overwritten  by identify_siblings()
808 	 * for Multi-Threading/Multi-Core capable CPUs
809 	 */
810 	c->threads_per_core = c->cores_per_socket = c->num_log = 1;
811 	c->socket_id = -1;
812 
813 	identify_siblings(c);
814 
815 	if (c->threads_per_core > smp_num_siblings)
816 		smp_num_siblings = c->threads_per_core;
817 #endif
818 	c->ppn = cpuid.field.ppn;
819 	c->number = cpuid.field.number;
820 	c->revision = cpuid.field.revision;
821 	c->model = cpuid.field.model;
822 	c->family = cpuid.field.family;
823 	c->archrev = cpuid.field.archrev;
824 	c->features = cpuid.field.features;
825 	c->model_name = get_model_name(c->family, c->model);
826 
827 	status = ia64_pal_vm_summary(&vm1, &vm2);
828 	if (status == PAL_STATUS_SUCCESS) {
829 		impl_va_msb = vm2.pal_vm_info_2_s.impl_va_msb;
830 		phys_addr_size = vm1.pal_vm_info_1_s.phys_add_size;
831 	}
832 	c->unimpl_va_mask = ~((7L<<61) | ((1L << (impl_va_msb + 1)) - 1));
833 	c->unimpl_pa_mask = ~((1L<<63) | ((1L << phys_addr_size) - 1));
834 }
835 
836 /*
837  * Do the following calculations:
838  *
839  * 1. the max. cache line size.
840  * 2. the minimum of the i-cache stride sizes for "flush_icache_range()".
841  * 3. the minimum of the cache stride sizes for "clflush_cache_range()".
842  */
843 static void
844 get_cache_info(void)
845 {
846 	unsigned long line_size, max = 1;
847 	unsigned long l, levels, unique_caches;
848 	pal_cache_config_info_t cci;
849 	long status;
850 
851         status = ia64_pal_cache_summary(&levels, &unique_caches);
852         if (status != 0) {
853                 printk(KERN_ERR "%s: ia64_pal_cache_summary() failed (status=%ld)\n",
854                        __func__, status);
855                 max = SMP_CACHE_BYTES;
856 		/* Safest setup for "flush_icache_range()" */
857 		ia64_i_cache_stride_shift = I_CACHE_STRIDE_SHIFT;
858 		/* Safest setup for "clflush_cache_range()" */
859 		ia64_cache_stride_shift = CACHE_STRIDE_SHIFT;
860 		goto out;
861         }
862 
863 	for (l = 0; l < levels; ++l) {
864 		/* cache_type (data_or_unified)=2 */
865 		status = ia64_pal_cache_config_info(l, 2, &cci);
866 		if (status != 0) {
867 			printk(KERN_ERR "%s: ia64_pal_cache_config_info"
868 				"(l=%lu, 2) failed (status=%ld)\n",
869 				__func__, l, status);
870 			max = SMP_CACHE_BYTES;
871 			/* The safest setup for "flush_icache_range()" */
872 			cci.pcci_stride = I_CACHE_STRIDE_SHIFT;
873 			/* The safest setup for "clflush_cache_range()" */
874 			ia64_cache_stride_shift = CACHE_STRIDE_SHIFT;
875 			cci.pcci_unified = 1;
876 		} else {
877 			if (cci.pcci_stride < ia64_cache_stride_shift)
878 				ia64_cache_stride_shift = cci.pcci_stride;
879 
880 			line_size = 1 << cci.pcci_line_size;
881 			if (line_size > max)
882 				max = line_size;
883 		}
884 
885 		if (!cci.pcci_unified) {
886 			/* cache_type (instruction)=1*/
887 			status = ia64_pal_cache_config_info(l, 1, &cci);
888 			if (status != 0) {
889 				printk(KERN_ERR "%s: ia64_pal_cache_config_info"
890 					"(l=%lu, 1) failed (status=%ld)\n",
891 					__func__, l, status);
892 				/* The safest setup for flush_icache_range() */
893 				cci.pcci_stride = I_CACHE_STRIDE_SHIFT;
894 			}
895 		}
896 		if (cci.pcci_stride < ia64_i_cache_stride_shift)
897 			ia64_i_cache_stride_shift = cci.pcci_stride;
898 	}
899   out:
900 	if (max > ia64_max_cacheline_size)
901 		ia64_max_cacheline_size = max;
902 }
903 
904 /*
905  * cpu_init() initializes state that is per-CPU.  This function acts
906  * as a 'CPU state barrier', nothing should get across.
907  */
908 void
909 cpu_init (void)
910 {
911 	extern void ia64_mmu_init(void *);
912 	static unsigned long max_num_phys_stacked = IA64_NUM_PHYS_STACK_REG;
913 	unsigned long num_phys_stacked;
914 	pal_vm_info_2_u_t vmi;
915 	unsigned int max_ctx;
916 	struct cpuinfo_ia64 *cpu_info;
917 	void *cpu_data;
918 
919 	cpu_data = per_cpu_init();
920 #ifdef CONFIG_SMP
921 	/*
922 	 * insert boot cpu into sibling and core mapes
923 	 * (must be done after per_cpu area is setup)
924 	 */
925 	if (smp_processor_id() == 0) {
926 		cpumask_set_cpu(0, &per_cpu(cpu_sibling_map, 0));
927 		cpumask_set_cpu(0, &cpu_core_map[0]);
928 	} else {
929 		/*
930 		 * Set ar.k3 so that assembly code in MCA handler can compute
931 		 * physical addresses of per cpu variables with a simple:
932 		 *   phys = ar.k3 + &per_cpu_var
933 		 * and the alt-dtlb-miss handler can set per-cpu mapping into
934 		 * the TLB when needed. head.S already did this for cpu0.
935 		 */
936 		ia64_set_kr(IA64_KR_PER_CPU_DATA,
937 			    ia64_tpa(cpu_data) - (long) __per_cpu_start);
938 	}
939 #endif
940 
941 	get_cache_info();
942 
943 	/*
944 	 * We can't pass "local_cpu_data" to identify_cpu() because we haven't called
945 	 * ia64_mmu_init() yet.  And we can't call ia64_mmu_init() first because it
946 	 * depends on the data returned by identify_cpu().  We break the dependency by
947 	 * accessing cpu_data() through the canonical per-CPU address.
948 	 */
949 	cpu_info = cpu_data + ((char *) &__ia64_per_cpu_var(ia64_cpu_info) - __per_cpu_start);
950 	identify_cpu(cpu_info);
951 
952 #ifdef CONFIG_MCKINLEY
953 	{
954 #		define FEATURE_SET 16
955 		struct ia64_pal_retval iprv;
956 
957 		if (cpu_info->family == 0x1f) {
958 			PAL_CALL_PHYS(iprv, PAL_PROC_GET_FEATURES, 0, FEATURE_SET, 0);
959 			if ((iprv.status == 0) && (iprv.v0 & 0x80) && (iprv.v2 & 0x80))
960 				PAL_CALL_PHYS(iprv, PAL_PROC_SET_FEATURES,
961 				              (iprv.v1 | 0x80), FEATURE_SET, 0);
962 		}
963 	}
964 #endif
965 
966 	/* Clear the stack memory reserved for pt_regs: */
967 	memset(task_pt_regs(current), 0, sizeof(struct pt_regs));
968 
969 	ia64_set_kr(IA64_KR_FPU_OWNER, 0);
970 
971 	/*
972 	 * Initialize the page-table base register to a global
973 	 * directory with all zeroes.  This ensure that we can handle
974 	 * TLB-misses to user address-space even before we created the
975 	 * first user address-space.  This may happen, e.g., due to
976 	 * aggressive use of lfetch.fault.
977 	 */
978 	ia64_set_kr(IA64_KR_PT_BASE, __pa(ia64_imva(empty_zero_page)));
979 
980 	/*
981 	 * Initialize default control register to defer speculative faults except
982 	 * for those arising from TLB misses, which are not deferred.  The
983 	 * kernel MUST NOT depend on a particular setting of these bits (in other words,
984 	 * the kernel must have recovery code for all speculative accesses).  Turn on
985 	 * dcr.lc as per recommendation by the architecture team.  Most IA-32 apps
986 	 * shouldn't be affected by this (moral: keep your ia32 locks aligned and you'll
987 	 * be fine).
988 	 */
989 	ia64_setreg(_IA64_REG_CR_DCR,  (  IA64_DCR_DP | IA64_DCR_DK | IA64_DCR_DX | IA64_DCR_DR
990 					| IA64_DCR_DA | IA64_DCR_DD | IA64_DCR_LC));
991 	atomic_inc(&init_mm.mm_count);
992 	current->active_mm = &init_mm;
993 	BUG_ON(current->mm);
994 
995 	ia64_mmu_init(ia64_imva(cpu_data));
996 	ia64_mca_cpu_init(ia64_imva(cpu_data));
997 
998 	/* Clear ITC to eliminate sched_clock() overflows in human time.  */
999 	ia64_set_itc(0);
1000 
1001 	/* disable all local interrupt sources: */
1002 	ia64_set_itv(1 << 16);
1003 	ia64_set_lrr0(1 << 16);
1004 	ia64_set_lrr1(1 << 16);
1005 	ia64_setreg(_IA64_REG_CR_PMV, 1 << 16);
1006 	ia64_setreg(_IA64_REG_CR_CMCV, 1 << 16);
1007 
1008 	/* clear TPR & XTP to enable all interrupt classes: */
1009 	ia64_setreg(_IA64_REG_CR_TPR, 0);
1010 
1011 	/* Clear any pending interrupts left by SAL/EFI */
1012 	while (ia64_get_ivr() != IA64_SPURIOUS_INT_VECTOR)
1013 		ia64_eoi();
1014 
1015 #ifdef CONFIG_SMP
1016 	normal_xtp();
1017 #endif
1018 
1019 	/* set ia64_ctx.max_rid to the maximum RID that is supported by all CPUs: */
1020 	if (ia64_pal_vm_summary(NULL, &vmi) == 0) {
1021 		max_ctx = (1U << (vmi.pal_vm_info_2_s.rid_size - 3)) - 1;
1022 		setup_ptcg_sem(vmi.pal_vm_info_2_s.max_purges, NPTCG_FROM_PAL);
1023 	} else {
1024 		printk(KERN_WARNING "cpu_init: PAL VM summary failed, assuming 18 RID bits\n");
1025 		max_ctx = (1U << 15) - 1;	/* use architected minimum */
1026 	}
1027 	while (max_ctx < ia64_ctx.max_ctx) {
1028 		unsigned int old = ia64_ctx.max_ctx;
1029 		if (cmpxchg(&ia64_ctx.max_ctx, old, max_ctx) == old)
1030 			break;
1031 	}
1032 
1033 	if (ia64_pal_rse_info(&num_phys_stacked, NULL) != 0) {
1034 		printk(KERN_WARNING "cpu_init: PAL RSE info failed; assuming 96 physical "
1035 		       "stacked regs\n");
1036 		num_phys_stacked = 96;
1037 	}
1038 	/* size of physical stacked register partition plus 8 bytes: */
1039 	if (num_phys_stacked > max_num_phys_stacked) {
1040 		ia64_patch_phys_stack_reg(num_phys_stacked*8 + 8);
1041 		max_num_phys_stacked = num_phys_stacked;
1042 	}
1043 	platform_cpu_init();
1044 }
1045 
1046 void __init
1047 check_bugs (void)
1048 {
1049 	ia64_patch_mckinley_e9((unsigned long) __start___mckinley_e9_bundles,
1050 			       (unsigned long) __end___mckinley_e9_bundles);
1051 }
1052 
1053 static int __init run_dmi_scan(void)
1054 {
1055 	dmi_scan_machine();
1056 	dmi_memdev_walk();
1057 	dmi_set_dump_stack_arch_desc();
1058 	return 0;
1059 }
1060 core_initcall(run_dmi_scan);
1061